"Radius" by BigBen2k

[Cathar]

Originally posted by MadDogMe
Cathar!, how thin would you have made your channels if you'd used a disc cutter/arbour?, how deep?(for which diam' blade?)...

What I mean is, do you think there's be much(any) improvement taking it down to .5mm?...

I really don't know. There's quite a number of things happening at once in the block I have, and it's not entirely intuitive with even a moderate understanding of what's going on.

As you drop down the channel width you'll also need to decrease the fin width. As you decrease the fin width you'll also find that you need less fin height. As you decrease the fin width and height you'll start to sacrifice structural integrity. You'll also have increased water resistance. Also the "play" of the nozzles will be affected quite significantly.

In short, it's not as simple as merely dropping the channel width and hoping it'll be better. There's a lot of complex interactions and geometry going on, including the all important structural considerations.

[MadDogMe]

Do you think there's a way of calculating all this?, should this be the thing to move towards?.
Where would you start?,predicted heatload?, then flowrate?, then predicted max achievable velocity?, then fin area needed , then optimum fin shape to meet area and maximise flow?.

I don't know!!... , I don't know enough yet to get my head around it!...

I'd like to see the theory mapped out tho' ...

[Cathar]

What you're basically asking is how to design a water-block scientifically.

It goes (roughly) like this:

1) Define target pump pressure
2) Derive optimal orifice size as a balance of pump pressure, water velocity, and volumetic flow rate.
3) Define target die size. Heat load is basically unimportant. Assume 100W (real watts - not fake Radiate watts) as a peak
4) Determine what material you're going to make the waterblock out of.
5) Derive structural integrity constraints to cope with up to 50kg of pressure over a 1cm^2 area given the material you've chosen.
6) Machine the material in a way that maximises the water surface area within 2mm of each side of the CPU die (ie. for a 10x10mm die, target a 14x14mm area to maximise surface area over)
7) Using 2) above, derive a method to maximise the water turbulence over the die area specifically - this will be greatly affected by the design you've chosen in 6)
8) After 7), derive the maximum height required of the water-block material and the base-plate thickness. These two are linked more strongly that one would first imagine. Determine this for your 100W heat load. Target the fin/channel/block height at the 98% dissipation mark. Meaning that 98% of the heat is being dissipated below that height. Basically this defines an upper limit to the useful height of the block. As a hint, for copper/100W/water, if you're focussing on anything above 8mm, you're going in the wrong direction, unless you have pathetically low flow rates/pump pressure.

After you've defined 8), you'll then learn that this impacts on 7), and almost always on 2), so you'll need to reiterate the process to refine it, while will undoubtedly bring 6) back into the mix as well, while keeping a very close eye on 5).

Keep refining and reiterating those points and you'll asymptotically approach the limitations of the waterblock's efficiency with the design architecture and machining method you've chosen.

I've been doing this reiteration process and managed to squeeze even more performance out of my design, but I'm also rapidly approaching the limitations of what can be done with the machining equipment I have access to.

[MadDogMe]

When you say 'derive optimal orifice size' do you mean the 'water area' of 'a cross section' of the block?(sorry about my terminology).
Is it best to have the block H2o area equal or slightly larger than tube bore?, for less 'back pressure' after the 'jetted' part?. or is the whole block slightly smaller so the whole thing is a 'high pressure' spot,(terminology?). I mean do you want to maintane the velocity all through the block or 'just' over the core. does it decrease overall pressure the longer the restriction is held(whole block) rather than just a restricted nozzle over core?. would it even work if the whole block was'nt restricted?.

My guess is the whole block but I have to ask ...

I thought max heatload would be relevant?. would thicker BP mean the fins overlapping the core by more than 2mm?(wider spread?). Is there a program like Radiate(I've not seen it yet) that shows you heat~spread but shows what H2o at various rates/speeds would remove as well?, to various shape/designs?. There should be! . Any thermal/fluid software you'd recomend?.

Thanx!, and thanx again!...

[myv65]

By "orifice size" what Cathar is referring to is the restriction he locates immediately above the fins. It's cross sectional area should generally be lower than that of the block's flow path. It's purpose to to increase the average velocity at the point where the water initially hits the block, which wonder-of-wonders occurs directly over the die.

The balancing act here is between higher average velocity in the "jet" vs higher total flow volume (higher average velocity every except the "jet"). With a "small" orifice, you'll get a higher peak water velocity which will yield a higher convection coefficient directly beneath the orifice. You'll also get lower total flow volume, meaning a reduced convection coefficient outside of the "jet". Vice versa for a "large" orifice.

[Cathar]

Thanks Dave for clarifying what I meant.

I thought max heatload would be relevant?. would thicker BP mean the fins overlapping the core by more than 2mm?(wider spread?). Is there a program like Radiate(I've not seen it yet) that shows you heat~spread but shows what H2o at various rates/speeds would remove as well?, to various shape/designs?. There should be! . Any thermal/fluid software you'd recomend?.

Max heatload is relevant. Like I said, assuming around a real 100W maximum would be fine. I was referring to it not being relevant to specifically target a block for some range of heat loads. Just pick a maximum and run with that. Every heat load less than your targetted maximum will then be catered for. The block won't behave "erratically" just because you throw a lower heat load at it, which was the point I was trying to make.

I've been packing boxes all day in preparation for moving house, so I was probably being too nit-picky on that point. My apologies.

I wrote a thermal simulator myself which is how I basically decided where to go with all this. Aesik has also written one which is a more advanced model than mine, and Les uses a mix of different web-based applications and some back-of-the-envelope calculations to pull it all together.

My program allows you to play with the convectional coefficient of the coolant and the flow rate, but doesn't allow you to vary it for different sections of the block.

[MadDogMe]

Do you calculate the inner diametre of the block to match the 1/2 tube though?, or does it not matter as long as it's not restricting flow(after the jet ) would making it larger help the jet in any way?. I can't imagine how to calculate flowspeed other than setting up a loop with the tube and rad and seeing what comes out the end, to the best 'squirt' I could get . the flow loss at sixfoot that was stated for a 1250 in the 'pump roundup' thread blew me away. did'nt realise there was so much impact ...

[bigben2k]

Lots of good comments and suggestions!

Fixittt: I don't get the need to create a similarly patterned block for chipset, VC, etc, other than the "neato" factor. I do like the idea of coming up with a Pro/Cooling kit, that someone (Danny) could make and sell, UNDER LICENSE! Even if we only get a buck for each kit sold, it might be a good funding source to get something started here. It'd be neat if Pro/Cooling sold this kit itself. Right now, the Pro/Gear section is only about links to Pro/Cooling logo stuff. Joe would have to chime in here, he's the bossman.

Gone_fishin: I'm too worried about structural integrity, to use a clear rubber sheet in this block. Remember, the fins, in combination with the top is the only thing holding this block together. If the baseplate starts to cave in, even by a fraction of a millimiter, it's dead.


I've been trying to re-create Les' thermal properties calculations, on Cathar's block, but I just can't figure it out. I'm gonna need Les in here to explain it to me!

I'll be giving another thermal calculator a shot. Will report later.

[myv65]

Originally posted by bigben2k
I'm too worried about structural integrity, to use a clear rubber sheet in this block. Remember, the fins, in combination with the top is the only thing holding this block together. If the baseplate starts to cave in, even by a fraction of a millimiter, it's dead.

So what does this mean? Are you planning on an o-ring (what I recall from earlier discussion)? Silicone RTV?

Anything you put in the interface will have some finite thickness. What you need to do is design a finite and tightly controlled gap between the top and base to exist when the top of the fins touches the bottom side of the top. Then you need to fill this gap with a material such that you can compress it to fill the gap while developing the right pressure to seal the interface yet not so high as to bow anything.

A solid o-ring? This can be a good choice but you need to use a relatively soft material and calculate the proper compression. Same goes for a "sheet" gasket. The most idiot-proof (no insult intended) manner is RTV because the excess will easily squeeze out if assembled right after application. It's sort of self-limiting in terms of the interface pressure it will let you develop.

This really isn't a hard problem, but isn't one that you should merely ignore, either. If statics isn't your thing, there's plenty of cookbook solutions to beam loading situations that'll help you see the orders of magnitude involved with deflection vs geometry and loading.

[gone_fishin]

Originally posted by bigben2k
Lots of good comments and suggestions!

Fixittt: I don't get the need to create a similarly patterned block for chipset, VC, etc, other than the "neato" factor. I do like the idea of coming up with a Pro/Cooling kit, that someone (Danny) could make and sell, UNDER LICENSE! Even if we only get a buck for each kit sold, it might be a good funding source to get something started here. It'd be neat if Pro/Cooling sold this kit itself. Right now, the Pro/Gear section is only about links to Pro/Cooling logo stuff. Joe would have to chime in here, he's the bossman.

Gone_fishin: I'm too worried about structural integrity, to use a clear rubber sheet in this block. Remember, the fins, in combination with the top is the only thing holding this block together. If the baseplate starts to cave in, even by a fraction of a millimiter, it's dead.


I've been trying to re-create Les' thermal properties calculations, on Cathar's block, but I just can't figure it out. I'm gonna need Les in here to explain it to me!

I'll be giving another thermal calculator a shot. Will report later.

And if the O-ring holds the plates apart by a fraction of a millimeter, then what? There will be nothing but open space in the gap for a fractional flex to happen. In the o-ring configurations I have seen, there is always a slight gap left. Either way a structural solution must be built into the top plate if this is your biggest concern.

[myv65]

If done right, you would have a gap immediately surrounding the o-ring. You would not have a gap where the top met the fins. In this situation, you need enough compression on the o-ring to form a water-tight seal but no more. You basically want to crunch the o-ring until the top hits the fins and that's it. At this point, the pressure from compression must seal the joint. If you continue to tighten further, you'll begin to bow the top down at the edges (outside the ring) and up in the center (over the fins). The larger the o-ring diameter, the more "forgiving" the junction gets. Softer o-rings are also obviously more compliant, but this means less interface pressure for a given deflection. It isn't exactly rocket science stuff, but you do need to understand how it all works to pull it off properly. Hollow o-rings are often effective because they offer decent resistance over a relatively wide deflection range up until the center is fully compressed.

[gone_fishin]

Having a cylinder inside a cylinder with an o-ring between the sidewalls is the proper use of an o-ring, not to mate two flat surfaces.

[Dix Dogfight]

How about something like this to get structural support in the middle?
(oooohhhhh paint skillz )

[MadDogMe]

I don't get the need to create a similarly patterned block for chipset, VC, etc, other than the "neato" factor.

Remember it's to be the ultimate waterblock design, whats good for the Ganders good for the northbridge ...
PS, is'nt that the whole point of milling a channel for the O ring?, so it allows the two surfaces to mate?, they don't go to that trouble(expensce) to stop it slipping around do they?...

[Cathar]

I think this whole O-ring thing is being blown slightly out of proportion here.

I use two O-rings to seal my block. I clamp it down tight. I stuck some thermal paste on the base of the middle plate as a means of checking and there is good contact being made between the tops of the fins and the bracing plate.

[gone_fishin]

Originally posted by Cathar
I think this whole O-ring thing is being blown slightly out of proportion here.

I use two O-rings to seal my block. I clamp it down tight. I stuck some thermal paste on the base of the middle plate as a means of checking and there is good contact being made between the tops of the fins and the bracing plate.

How thick was the paste? Now you must rely on the purchaser to retighten it properly in order for it to not have a gap and flex? You know the first thing anyone is going to do once they get their hands on a sectional block is to take the darn thing apart.


Edit: Cathar, I just realized you said you use two. This means that you use one inbetweeen your top and middle plate. Now I think I've seen the pic you shown somewhere else that the middle plate has a slot of some type to induce an impingement but this also applies a back presure. The other two exit holes in the center and top plate have what blocking the backpresured water from going straight to the outlets before ever entering the block (A gasket in this top level would eliminate that problem and boost impingment because this escape route for the water would be eliminated. Now tell me you don't like gaskets after a free improvement like that) ?

[bigben2k]

Hum... this is going around in circles!

IMO, the groove for an o-ring is the best option, but I will have to consider a silicone sealer. Here's why:

Water will respond very nicely to an o-ring, without a silicone sealer. In my rig, phase 1 is a straight watercooler, in it's simplest form. Phase 2 involves adding a TEC. Phase 3 adds a water chiller. When I go to phase 3, I'll be using a water/methanol mixture.

With methanol, an o-ring doesn't work as well, because the liquid is so thin, that it will easily bypass the o-ring.


Dix Dogfight: I'm gonna refer you to the link to Paul Vodrazka's info (see previous post). The compression on the o-ring is minimal. The fin pattern is flush with the top of the bottom piece of this block, so everything should clamp down nicely with a flat top.

Got an e-mail from Utabintarbo this morning. I think I passed my paranoia to him! The copper top might become a requirement, so unless myv65 can point us to a mechanical link, and I can calculate that the plexi/acrylic top is viable, it's gonna have to be copper :shrug: .

Going over AMD specs, it seems that the four corner bolt isn't appropriate (flash!). It looks like the mounting pressure must be applied longitudinally across the core. I don't get it though, what keeps the block from tipping in the other direction, crushing the core? The little pads? Are they soft rubber, or do they have any kind of solidity to them?

Here's a pic (not for Radius).

In this configuration, we could add tabs to the sides of the block.

[Cathar]

Originally posted by gone_fishin
Cathar, I just realized you said you use two. This means that you use one inbetweeen your top and middle plate. Now I think I've seen the pic you shown somewhere else that the middle plate has a slot of some type to induce an impingement but this also applies a back presure. The other two exit holes in the center and top plate have what blocking the backpresured water from going straight to the outlets before ever entering the block (A gasket in this top level would eliminate that problem and boost impingment because this escape route for the water would be eliminated. Now tell me you don't like gaskets after a free improvement like that) ?

Okay, just to prove the point.

I assembled a block without the barbs in the top plate so we can easily see what's going on. I then snapped a piccy in macro mode with my digicam focusses on the central barb area where the top plate meets the middle plate. The O-rings are both in place on this block, just no barbs so we can see what's going on. Here's what I snapped:



Look at that picture for a bit. I've outlined where we can clearly see that there's no gap on the left. On the right the barb thread casts a shadow from the angle of the flash. I shone a bright torch at a very shallow angle to look underneath there on the right middle and it really is just a shadow.

Now, with whatever gap you imagine that you can see there pales in comparison to the channels which can be seen in the bottom middle of the picture. The inlet nozzle is about 3/4 the size of the channels effectively, and even then I can't see how any water of any significant volume could possibly be forcing its way between the middle and top plates.

Like I said, this O-ring thing is being blown out of proportion. I simply cannot see a gap. If one exists, then it's going to range in the microns in size, compared to the millimeters in size of the inlet nozzle.

[utabintarbo]

Originally posted by bigben2k

Going over AMD specs, it seems that the four corner bolt isn't appropriate (flash!). It looks like the mounting pressure must be applied longitudinally across the core. I don't get it though, what keeps the block from tipping in the other direction, crushing the core? The little pads? Are they soft rubber, or do they have any kind of solidity to them?

Mounting proposal:

Central boom with outriggers. Might solve the "balance" question?

[myv65]

Originally posted by bigben2k
With methanol, an o-ring doesn't work as well, because the liquid is so thin, that it will easily bypass the o-ring.

unless myv65 can point us to a mechanical link, and I can calculate that the plexi/acrylic top is viable, it's gonna have to be copper

On the first point, not really. I'd be more concerned that you choose a material that's compatible with methanol. It's pretty easy to develop 10 psig compression on the o-ring (or a heckuva lot more). Doesn't much matter what the viscosity is provided you have good surface finish, a compliant o-ring, and sufficient pressure.

On the second point, I always go to Rourk's Formulas for Stress and Strain, by Warren C. Young. I've got a copy of the sixth edition. Not sure what edition is current. Be warned, however, that it darn near takes a full ME degree to make any sense of the copious tables in this book. Also note that you need to have a basic understanding of superposition (adding up the individual load cases to arrive at your "composite" load case).

[bigben2k]

Hum... I think that a combination of a silicone sealer, along with an o-ring would be best. Industrially, they use huge mounting flanges to connect pipes, that seem way oversized.

Calculating the structural load is out of my league. I give up! I'll have to take my chances with a thick piece of acrylic/perspex/whatever-other-names-are-out-there clear material.

In my search though, I found an interesting link:
http://www.engr.usask.ca/~drs694/fluidmechanics.htm

Utabintarbo: I think I like your idea best. I just wish I knew more about those 4 pads on the Athlons. Are they just common rubber?

As per AMD specs, the core is between 0.80 and 0.88 mm high, off the face of the processor (so that's why we don't use a shim!). The pads are max 1.90 mm thick, so I have to assume that they have a certain "sponginess".

Otherwise, the whole processor width (it is square) will fall within 49.27 mm and 49.78 mm. The core is either 7.31 by 11.06 (for Model 8 with CPUID of 681) or 7.47 by 11.33 mm (for Model 8 with CPUID of 680). Now which is Thoroughbred rev B?

I'll be back with socket specs.

[bigben2k]

Here's the latest draw (thanks to Utabintarbo)

Note that the first set of fins are now cut right through.

(I think that the fin pattern is more than 30mm?!?)

[myv65]

Originally posted by bigben2k
Hum... I think that a combination of a silicone sealer, along with an o-ring would be best. Industrially, they use huge mounting flanges to connect pipes, that seem way oversized.

Industrial flanges in the US are normally rated in "pounds", with 150 and 300 pound most common. This refers to the pressure rating with steam as the medium. For obvious reasons (to an ME geek, anyway) these are conservative ratings that guarantee against a failure under the rated steam pressure. At the turn of the previous century, boiler explosions were a common occurence. Thanks to ASME codes, they are very, very uncommon today.

Anyway, that's the reason for the heavy flanges in "industry".

[gone_fishin]

Originally posted by Cathar
Okay, just to prove the point.

I assembled a block without the barbs in the top plate so we can easily see what's going on. I then snapped a piccy in macro mode with my digicam focusses on the central barb area where the top plate meets the middle plate. The O-rings are both in place on this block, just no barbs so we can see what's going on.
Look at that picture for a bit. I've outlined where we can clearly see that there's no gap on the left. On the right the barb thread casts a shadow from the angle of the flash. I shone a bright torch at a very shallow angle to look underneath there on the right middle and it really is just a shadow.

Now, with whatever gap you imagine that you can see there pales in comparison to the channels which can be seen in the bottom middle of the picture. The inlet nozzle is about 3/4 the size of the channels effectively, and even then I can't see how any water of any significant volume could possibly be forcing its way between the middle and top plates.

Like I said, this O-ring thing is being blown out of proportion. I simply cannot see a gap. If one exists, then it's going to range in the microns in size, compared to the millimeters in size of the inlet nozzle.

Nice pic. You have it tightened appropriately. Are you confident that users will? If they do not then you will loose force through the nozzle and get reports of lower performance. All I am suggesting is a simple gasket will eliminate that possibillity. It will also cut down on an extra milling step and the cost of a gasket is equal to the cost of an o-ring. I'm not critisizing, just trying to help.

[BillA]

I'm guessing that if myv65 can mention Roark, anything goes !
(an ind std ref BTW, a good read but tough sledding for the untutored)

after designing and doing failure analysis of o-ring joints for many years (in both static and cyclic pressure service),
there are several useful guidlines:
- when the component surfaces make contact, as they must, there should be 30 to 40% compression of the crossection
- the o-ring's volume should be between 60 and 70% of the groove's
- using a 30 to 50 Shore A durometer (hardness) material will make the surface finish less critical
- the o-ring's (centerline) length should match the groove's (do NOT stretch)
- the groove must have an outer containment ring to 'capture' the o-ring, else it is functioning as a gasket (different design basis)

do not use a sealant with the o-ring (no need if properly designed, will overfill the groove, make disassembly a pain, etc)

g_f
o-rings, properly designed, are FAR more reliable than gaskets